darwin: workaround PROC_PIDLISTFDS on processes with no fds

This has been causing various seemingly spurious CI failures as well as
some failures on people running tests on beta builds.

lix> ++(nix-collect-garbage-dry-run.sh:20) nix-store --gc --print-dead
lix> ++(nix-collect-garbage-dry-run.sh:20) wc -l
lix> finding garbage collector roots...
lix> error: Listing pid 87261 file descriptors: Undefined error: 0

There is no real way to write a proper test for this, other than to
start a process like the following:

int main(void) {
    for (int i = 0; i < 1000; ++i) {
        close(i);
    }
    sleep(10000);
}

and then let Lix's gc look at it.

I have a relatively high confidence this *will* fix the problem since I
have manually confirmed the behaviour of the libproc call is
as-unexpected, and it would perfectly explain the observed symptom.

Fixes: https://git.lix.systems/lix-project/lix/issues/446
Change-Id: I67669b98377af17895644b3bafdf42fc33abd076

2 files changed, 31 insertions, 1 deletions

diff --git a/doc/manual/rl-next/haunted-gc-macos.md b/doc/manual/rl-next/haunted-gc-macos.md
new file mode 100644
index 000000000..3ce912b2d
--- /dev/null
+++ b/doc/manual/rl-next/haunted-gc-macos.md
@@ -0,0 +1,15 @@
+---
+synopsis: "Fix unexpectedly-successful GC failures on macOS"
+cls: 1723
+issues: fj#446
+credits: jade
+category: Fixes
+---
+
+Has the following happened to you on macOS? This failure has been successfully eliminated, thanks to our successful deployment of advanced successful-failure detection technology (it's just `if (failed && errno == 0)`. Patent pending<sup>not really</sup>):
+
+```
+$ nix-store --gc --print-dead
+finding garbage collector roots...
+error: Listing pid 87261 file descriptors: Undefined error: 0
+```
diff --git a/src/libstore/platform/darwin.cc b/src/libstore/platform/darwin.cc
index 1b591fde3..1f7e9be23 100644
--- a/src/libstore/platform/darwin.cc
+++ b/src/libstore/platform/darwin.cc
@@ -56,12 +56,27 @@ void DarwinLocalStore::findPlatformRoots(UncheckedRoots & unchecked)
             while (fdBufSize > fds.size() * sizeof(struct proc_fdinfo)) {
                 // Reserve some extra size so we don't fail too much
                 fds.resize((fdBufSize + fdBufSize / 8) / sizeof(struct proc_fdinfo));
+                errno = 0;
                 fdBufSize = proc_pidinfo(
                     pid, PROC_PIDLISTFDS, 0, fds.data(), fds.size() * sizeof(struct proc_fdinfo)
                 );
 
+                // errno == 0???! Yes, seriously. This is because macOS has a
+                // broken syscall wrapper for proc_pidinfo that has no way of
+                // dealing with the system call successfully returning 0. It
+                // takes the -1 error result from the errno-setting syscall
+                // wrapper and turns it into a 0 result. But what if the system
+                // call actually returns 0? Then you get an errno of success.
+                //
+                // https://github.com/apple-opensource/xnu/blob/4f43d4276fc6a87f2461a3ab18287e4a2e5a1cc0/libsyscall/wrappers/libproc/libproc.c#L100-L110
+                // https://git.lix.systems/lix-project/lix/issues/446#issuecomment-5483
+                // FB14695751
                 if (fdBufSize <= 0) {
-                    throw SysError("Listing pid %1% file descriptors", pid);
+                    if (errno == 0) {
+                        break;
+                    } else {
+                        throw SysError("Listing pid %1% file descriptors", pid);
+                    }
                 }
             }
             fds.resize(fdBufSize / sizeof(struct proc_fdinfo));